Work

Physics-informed generative real-time lens-free imaging

Year: 2025

Type: article

Abstract: High-throughput biomedical applications increasingly demand real-time, wide field-of-view (FOV) imaging. Conventional 2D lens-free imaging (LFI) systems expand the FOV but are often limited by slow multi-position measurements, complex data pre-processing, and tight optical constraints, making them suitable only for static or thin samples. Here we report GenLFI, a real-time imaging framework that combines a large-FOV LFI setup with a generative, unsupervised physics-informed neural network (PINN) for direct holographic reconstruction without the need for multiple measurements. GenLFI enables robust 2D imaging of 3D dynamic samples, such as droplet-based microfluidics and 3D cell cultures, in complex optical environments. Unlike previous methods, our approach removes dependency on optical setup parameters, allowing a FOV limited only by hardware. We demonstrate a real-time FOV exceeding 550 mm 2 , more than 20 times larger than current systems. GenLFI provides a scalable and practical solution for advancing automation and throughput in biomedical imaging workflows.

Source: Cell Reports Physical Science

DOI: 10.1016/j.xcrp.2025.102792